Forwarded From: Robert Hettinga <rah@shipwright.com>
Originally From: Dan Geer <geer@world.std.com>
-------------------------------------
Risk Management is Where the Money Is
-------------------------------------
Digital Commerce Society of Boston
3 November 98
Daniel E. Geer, Jr., Sc.D.
Senior Strategist, CertCo, Inc.
55 Broad Street, NYC, and
100 Cambridgepark Drive, Cambridge
geer@certco.com
Given my biases, I am going to describe where the future of the security
marketplace is and where it is not. I will argue that the financial
community is and remains the place to look for "first light" for new
security technology. I will give you a rundown of what's new while I
predict what little time is left for many of today's products, purveyors
and regulators. I will argue that, in many ways, the party's over for the
security field as we know it now. I will range broadly because security,
as a concept, is universal.
"Nothing is so powerful as an idea whose time has come." For security
technology, that time is now. IBM calls the three requirements of the
"e-business" future as: #1 security, #2 scalability, and #3 integration.
Forrester, Gartner, META, Yankee and all the other analysts agree -- the
most important enabling technology for electronic business, besides
network connectivity itself, is security. AD Little estimates that
security, privacy and the legal issues of digital signature together
constitute over half of the quantifiable barriers to electronic commerce.
There are whole venture funds whose investment focus is around security.
Security startups are everywhere; so are security books. The word
"security" is hardly rare in employment advertisements. You cannot walk a
trade show and not see the word "security" in screaming big type. The
number of security meetings is preposterous. Presidential Commissions are
busy spending real money on security for the information systems that run
the country.
"In the future, everyone will each get 15 minutes of fame." That applies
to security,too. Today's security specialty companies cannot all survive;
they can be eclipsed by the platform vendors too easily. Only platform
vendors can deliver security that is integrated enough to scale and
invisible enough to ignore. Even the Justice Department knows that once
something is in the operating system, any independent market for it
collapses. Yes, security's time may well have come, but in a Warhol world,
that would mean that it is about time to go.
The focus of "security" research today is the study of "trust management"
-- how trust is defined, created, annotated, propagated, circumscribed,
stored, exchanged, accounted for, recalled and adjudicated in our
electronic world. This is natural because security is a means and not an
end. This is mature because all technology differentiates along
cost-benefit lines. All the security technology that you can buy today
enables some aspect of trust management and novel variations show up
daily.
You can walk out of this hall and buy systems that use passwords that get
local machines to trust you enough to let you in. You can buy smart cards
that can do your cryptographic calculations for you, respond to
challenges, hold your keys inviolable or, more interestingly, have
identities of their own and serve merely to introduce you on their own
terms. You can buy biometric devices that look at your voice, your face,
your retina, your fingerprint, or even the idiosyncrasies of how you
learned to type and so say, "Yep, that's the guy." You can get systems
that are sufficiently hardened that you can rely on them if for no other
reason they are so nearly useless no one would want to break in. You can
still get your hands on security systems in the raw and roll your own
directly from source-code. You can, anywhere, anytime, spin-up virtual
private networks that are trustworthy protectors of your confidentiality
however hostile the intervening wires are. You can even deliver privacy
between strangers -- nearly a matter of creating trust in order to
propagate it. You can put a document into the Eternity Service and trust
that it can never be erased or you can put it into a cryptographic file
system and trust that it can never be found. Simple? Yes; academics and
entrepreneurs alike are busy supplying ways to propagate trust.
They have it all wrong.
If you ever took a course in probability then you know that many problems
are solved by calculating their dual -- the probability of "not X" can be
a whole lot more tractable than figuring Pr(X) directly. If you're in a
security-based startup company, then you'll know that making money
requires making excitement, even if the excitement is somebody else's
public humiliation. And all of you can agree that the more important
something is, the more it must be managed. Trust management is surely
exciting, but like most exciting ideas it is unimportant. What is
important is risk management, the sister, the dual of trust management.
And because risk management makes money, it drives the security world from
here on out.
Every financial firm of any substance has a formal Risk Management
Department that consumes a lion's share of the corporate IT budget. The
financial world in its entirety is about packaging risk so that it can be
bought and sold, i.e., so that risk can be securitized and finely enough
graded to be managed at a profit. Everything from the lowly car loan to
the most exotic derivative security is a risk-reward tradeoff. Don't for a
minute underestimate the amount of money to be made on Wall Street, London
and/or Tokyo when you can invent a new way to package risk. The impact of
Moore's Law on the financial world is inestimable -- computing has made
that world rich because it has enabled risk packaging to grow ever more
precise, ever more real-time, ever more differentiated, ever more
manageable. You don't have to understand forward swaptions, collateralized
mortgage obligations, yield burning, or anything else to understand that
risk management is where the money is. In a capitalist world, if something
is where the money is, that something rules. Risk is that something.
Security technology has heretofore been about moving trust around as if
risk is definitionally undesirable and reliable trust management simply
obviates the issue of risk. It does not come close. In two years time the
"trust-hauling" market will be somewhere on the down-slope between legacy
and dead. Risk management is going to take over as the dominant paradigm
because risk management can subsume trust, but trust management cannot
subsume risk. The Internet has made this so.
The Internet is irresistible because it lowers barriers to entry on a
global basis -- global in both space and time. Ever more important parts
of the world's economy exist only in cyberspace, and lead times have
entirely collapsed. Every professional fortune teller is bidding
geometric increases in the dollar volume of electronic commercial
activity. But when there is enough booty available, even absurdly
difficult attacks become plausible. This is the world we are in. It will
never be possible to really do the job of trust management any more than
it is possible to really win an arms race or really preclude your car from
being stolen. But risk management -- that is doable and it is doable at a
profit. The proof is all around us.
We are a score of years down this road. 1978 was a vintage security year;
the remarkable papers by Rivest, Shamir & Adleman and Needham & Schroeder
were published, both in CACM as it happens. The former introduced public
key ideas and the latter created Kerberos. The counterpoint between these
two technologies is instructive. Both symmetric cryptosystems, like
Kerberos, and asymmetric cryptosystems, like RSA, do the same thing --
that is to say they do key distribution -- but the semantics are quite
different. The fundamental security-enabling activity of a secret key
system is to issue fresh keys at low latency and on demand. The
fundamental security-enabling activity of an asymmetric key system is to
verify the as-yet-unrevoked status of a key already in circulation, again
with low latency and on demand. This is key management and it is a systems
cost; a secret key system like Kerberos has incurred nearly all its costs
by the moment of key issuance. By contrast, a public key system incurs
nearly all its costs with respect to key revocation. Hence, a rule of
thumb: The cost of key issuance plus the cost of key revocation is a
constant, just yet another version of "You can pay me now or you can pay
me later."
Because of the tradeoffs between who pays for what part of the systems
cost and who gets the benefit, secret key systems and public key systems
have different fields of use. Secret key systems are fast and offer
revocation at no marginal cost. Public key systems are slow but they
enable digital signature and thus enable proof of action, non-repudiation
as it is called. Secret key systems are the default choice within an
organization while public key systems are the default choice between
organizations, i.e., secret key for where security is an intramural
concern intramurally arbitrated, and public key for where security is
extramural thereby requiring recourse to a third party judge in cases of
dispute. The relentless blurring of what is intramural and what is
extramural will favor public key over time.
Because a trust management paradigm says that a digital signature is only
as valid as the key (in which it was signed) was at the moment of
signature, it is only as good as the procedural perfection of the
certificate issuer and the timely transmission of any subsequent
revocation. **These are high costs.** In fact, the true costs of general
public key infrastructure are so extraordinarily high that only our
collective ignorance of those costs permits us to propel ourselves toward
a general PKI as if it were a panacea. When, not if, the user community
at large realizes this, we "security people" will have but two choices,
compromise on (gloss over) the quality of trust that public key can
deliver or back off from the claims of full trust cheap. In other words,
we'll have to fit the benefit to the endurable cost or fit the cost to the
requisite benefit. Since, as a rule of thumb, to halve the probability of
loss you have to at least double the cost of countermeasures, any finite
tolerance of cost means an upper bound on how much security you can get.
In the fullness of time, security technology will be evaluated on the same
cost-benefit-risk tradeoff on which other technologies are evaluated. This
is the price of maturity; this is the price not yet paid.
Do not misunderstand me; public key technology, secret key technology,
security technology in general are daily reaching new levels of protective
capability. What they cannot protect against is being over-sold, and they
are being over-sold. Why is that?
The days when the Internet was a toy are gone even if a high percentage of
its new investors are still coming in merely to avoid looking dowdy. The
real question on the table is: When does the Internet become more like the
data center. And what does making the Internet more like the data center
mean? At a minimum, it means metered use. Discussions are already
widespread about requiring Internet postage; large ISPs will probably
demand it, existing postal services would love to sell it and data
centers, such as the financial giants, will get a better handle on what
goes in and out the door. At least one Wall Street bank already does
charge-back for network bandwidth consumption and their internal
electronic security regime plays a role in assigning those costs just as,
in turn, their security group manages the user database via incremental
updates rather than fresh full copies so as to minimize their bandwidth
charges. That's not postage, but it is close and it is now.
Incremental use charges are but one example, interesting mostly because
they are a near term step toward making the Internet into a data center.
The fundamental value of the data center is the information it holds. The
past few years have seen data warehousing, data mining and now connection
of the data center to the Web, data publishing if you will. MVS, for
example, has a really good web server and someone in the audience will
have to convince me that there is a difference between a 1970's central
time-share machine and an MVS web server in a swarm of "thin clients" on
fast networks. It certainly isn't the direct wire connection -- SSL
simulates that well enough. It surely isn't the management model; the MIS
director who had declared defeat in desktop configuration management will,
you can be sure, rejoice at getting control back.
In the mainframe world, you move the computation to where the data is. In
a client server world, you move the data to where the computation is. Web
servers front-ending corporate databases attached to virtual private
networks full of some universal client like a web browser sure sounds like
a resurgence of the data center to me. The IBM 390 is a good machine and
the Wintel cartel has pretty much ensured that no upstart will enter their
space. From Wintel's point of view, using all those desktop cycles for
display functions is just fine. Could it be that simple?
Financial markets made SUN what it is today and vice versa -- SUN's first
big win, the first big demonstration that computing power had risen to
such a degree that moving the data to where the computing is made sense,
"the network is the computer" and all that. Financial markets, in the
sense of traders going head to head, used that power to replace whom you
knew with what you know and set off a technology-as-weapon metaphor that
has overtaken most of the business world. Financial Markets, in the sense
of Exchanges, now rely on a dense spread of computing that exceeds what
most of us have to deal with; more than one major bank has 15,000 FTP jobs
a night just moving data to or from its data center. Plenty of staff at
the NYSE lose $1000 apiece for every 15 minutes the Exchange is late
opening due to IT unavailability. No computing equipment is too expensive
when trumped with "I can make that back on the first trade." No small
country runs its currency anymore.
There was once no question that the fundamental purpose of an exchange was
to provide "an advantage of time and place" to those who would trade on it
and, in so doing, establish efficiency and liquidity baselines against
which others would be judged. Beginning first with the "Paperwork Crisis"
in the 60's and reaching a crescendo after the "Crash of '87," the
Exchanges have been fully committed to electronic commerce before that
phrase meant anything. But since the Internet, time and place are
meaningless and the Exchanges know it. They are working hard to make
oversight, fair play and quality of service into new baselines. Clearly,
security technology is #1 in their list of requirements followed closely
by scalability and integration.
Security in a financial world market that is both nowhere and everywhere
is a difficult thing to define well enough to solve, but if there is
anything to engineering as a discipline then it is that the heavy work is
in getting the problem statement right. So, to return to my central
premise, if new security technology is a result of investment and if the
investment in security technology is naturally centered within the
financial community, what is the problem statement? ** If we get that
right, we can predict the future. **
I submit that the problem statement is how to bring a transactional
semantic to the Internet. This is not a new problem, but it is an as yet
unsolved one. The existing financial markets want transactions because
transactions are what they are about and transactions are what they know.
Upstarts like the payment vendors want to be the first to deliver
transactions and disintermediate the financial firms. Technical legal
beagles reason that there is no transaction without recourse, no recourse
without contract, no contract without non-repudiation and no
non-repudiation without digital signature. Anyone who wants to do
business on the Web needs transactions.
Hal Varian, an economist and Dean of the Information Management School at
Berkeley, taught me that what the Internet changes more than anything else
is that it brings the efficiency of auction to markets that never had that
option. This is a cover story in this week's "The Industry Standard."
Auctions need security technology because what makes an auction an auction
is the ability to conclude a transaction which, by its own execution,
"discovers" a price. In other words, the nature of the world's economy is
changed by the existence of the Internet, but only on the condition that
electronic transactions are up to job.
So what do I mean by "transaction?" I mean a non-repudiable communication
between two parties who can each verify the time-, value- and
content-integrity of that communication, who can presume confidentiality
of that communication, who can verify the authenticity and authorization
of their counterparty and who can present all these evidences to third
party adjudication should there be a need for recourse at any arbitrary
time in the future. **Every single part of that definition begs the
question of security mechanism.** It is on that basis I claim that the
security technology of tomorrow will be crafted in response to the unmet
needs of financial markets today.
As an example, your handwritten signature on a check is what, in
principle, authorizes that funds move from A to B. In truth, from a bank's
point of view, actually verifying handwritten signatures is a transaction
cost that is not worth bearing unless the cost of verification is less
than the risk of loss. At the largest banks, the threshold dollar amount
below which verification does not really happen is a closely guarded
number, but it generally exceeds $20,000 and still they have platoons of
people doing this all day, every day. Converting the means of signature
verification from a manual process into a machine-able one would radically
change the economics of check processing. It would add billions to lines
and do it from the cost-avoidance side of the ledger.
But that is not all. Some $300B of U.S. payments are made every day of
which only $60B are in the form of checks; the balance is largely in cash
transactions of $5 or less. From both the merchant's and the bank's
perspectives, getting rid of cash would be a huge win because handling
costs for small dollar amounts often exceed the profit margins on the
underlying sales. While the consumer may well adopt cashless payment out
of some sense of convenience, the financial side of the house will enable
it to avoid costs.
Only this morning, Frost & Sullivan released a study that defines
e-commerce as "commercial transactions taking place over the Internet with
exchange of value in real time." Web payment sparked numerous startups
with numerous different mechanisms. It is too late for you to enter this
market, but it is not too late for those payment-systems vendors to
rethink what they are trying to do. All of them are suffering because the
volume of Web-based retail business has not picked up as fast as their
business plans had presumed. For the retail customer, the main thing the
Web offers is product discovery; a good print catalog and an 800 number
are otherwise hard to beat. It is clear that the real money in Web
commerce is in business-to-business commerce, but there the supply chain
has a lot more complication and the kinds of security mechanisms need to
be better than those for buying a toaster oven. Whereas retail commerce
is about small dollar amounts and stranger-to-stranger transactions
through a financial intermediary like a credit-card company,
business-to-business is more about relationships, the dollar value of the
sale is much bigger, and banks play a direct role (through letters of
credit, collateralized bills of lading, etc.)
B2B commerce does not have a good solution yet. If you want to sell into
this market, be aware that the customer will buy either to avoid costs he
has now or to make revenue he doesn't have yet. In the case of saving
costs, you'll have to sell him the technology on a turnkey basis -- he
will not cut you into the transactional revenue stream. If you can really
show that your technology will make him revenue he did not have a chance
to make otherwise, you may be able to get a piece of the revenue stream,
but do not underestimate the cost-avoidance focus of big buyers and
sellers. As far out as 2005, over half the Internet-transactions will be
transactions converted from paper and credit/debit cards, not new
transactions. **When selling into a cost-averse market you automate rather
than revolutionize, and you do not get a piece of the action.**
Everyone likes to talk about "disintermediating the banks," that is making
the intermediary role of banks in commerce less essential by performing
that service in some other way. Bill Gates is widely quoted as saying that
"Banks are dinosaurs." At the highest end, they are not dinosaurs and they
are not about to be disintermediated. Whilst the banks have a natural
affection for their income streams, that doesn't prevent
disintermediation. Most wiseguys trying to disintermediate the banks
misunderstand what banks do. This is what they do: They interpose their
balance sheet between the expectations of the counterparties to a
transaction and the risk of default on either of their parts. They
undertake stop-loss protections against credit risk, insolvency,
operational failure, currency fluctuation, diversion of funds delivery,
etc. In other words, they manage risk because they can absorb loss.
**Electronic commerce payment technology cannot absorb loss, so it cannot
and will not disintermediate the banks.**
Think of this this way: All public key technology is driven to make a
digital signature verifiable, i.e., it is about quality control and
guarantee on the signature itself. This is a stunning thing, but it is not
the whole equation. The intermediation role that banks play is to
guarantee the transaction, i.e., it is broader than just the verification
of a signature. The bank's know-how and its balance sheet are not
something that can be replaced by a cryptographic calculation. The
ability to avoid loss never makes up for the ability to absorb loss. The
cryptography guarantees the signature; the bank's capital guarantees the
transaction. **Risk control encapsulates trust.**
In the midst of this, you might say "What are the standards?" in the sense
of "What do the formal standards groups have to say?" The banking world is
regulation rich and standards rich, too, which begs the question -- "Which
standards matter?" The world of the Internet is making some of the
banking-centric standards passe' but, unlike the combination of standards
and regulations the banks are familiar with, the standards groups of the
Internet cannot take on accountability for the implications of
conformance/non-conformance though they continue to define it for others.
This makes Internet standards substantially difficult to swallow because
there is no accountability, nor can there be. The absence of enforcement
guarantees that the only Internet standards that will really get attention
are those that promote interoperability across jurisdictional boundaries.
Ironically, this is all the pioneers of the Internet ever wanted.
What the banks want, and I assure you they will get, is a set of
cryptographically sophisticated tools that move the risks of the Internet
from open-ended to estimable. In a sense, this is like insurability. It is
probably apocryphal, but the story goes that a major investment firm with
a Web commerce idea went to a big insurance company to seek stop loss
protection. The conversation supposedly went like this:
"How big is the potential loss?"
"We don't know."
"How likely is a loss to occur?"
"We don't know."
"How much is your company worth?"
"This much."
"That's the premium; send it in."
Whether true or not, it illustrates the point -- the issue is getting a
handle on the risk such that it can be priced. Every one of you who has
tried to sell security technology has discovered that the only willing
customers are those who either (1) have just been embarrassed in public or
(2) have just learned that they are facing an audit. Everyone else is an
unwilling customer. We've been dumb about this; we've tried to sell
security as a means to establish trust but we've done it by railing about
threats. It's no damned wonder that we haven't sold much. I know I have
often wondered if my market might not explode were I to get just one of
the big loss-prevention insurers to make good security practices and
technology into an underwriting standard. Then, just like "Do you have
sprinklers?" everyone is forced to confront whether they want to pay for
security or pay for non-security. I am confident that the insurers could
soften up my targets a lot better than I can.
Let me tell you, they are about to. Insurability of Web commerce is
essential, and no insurer is going to accept "We don't know" as an answer.
They will say "Send it all in" and they'll mean it. The demand side for
security technology is exploding but it isn't quite the security
technology we have on hand.
If a digital signature has the uniquely irreplaceable property of
providing proof to a judge, then the role of a "trusted third party" is
going to become more important over time, not less. Think of it this way:
when I get a certificate issued to me by a certifying authority, I do have
some risk around whether the CA is well operated or not. This includes the
probability they will issue a certificate with my public key but someone
else's name and whether when I tell them that my key has been compromised
they will spring into decisive action. Most of that risk I can handle by a
combination of due diligence and contract.
However, when I give my certificate to you and say "Hi, I'm here from
Central Services to fix your system" it is you that's in a risky position.
You have to say "Is this certificate valid?" That means you have to check
that the certificate is not listed as revoked, that the signature on the
certificate is well formed, that the certificate authority which issued
this certificate itself has an identity certificate that is itself validly
signed, that the certificate authority is itself not in any trouble with
revocation, and and so forth, ** recursively. **
The full cost of revocation testing is proportional to the square of the
depth of the issuance hierarchy. In other words, this exceeds the
intellectual capacity of most certificate recipients. This means that most
recipients cannot themselves rely on the security technology to establish
trust beyond the shadow of doubt. Instead, if recipients are smart, they
will turn again to the insurance world just as risk holders have done
whenever they cannot afford to carry on their books the consequences of a
remotely unlikely event. For the insurer, he will underwrite a guarantee
on the transaction for a fee that will reflect his experience with the
CA's practices, the kind of transaction undertaken, the dollar amounts
involved, etc. This will seem sensible to all parties because it is so
familiar. This is risk management underwritten by financial
intermediaries. This is where we will shortly be. This is the card eight
major banks and CertCo played ten days ago -- the formation of "a global
network of compliant businesses that use a common risk management
framework." **This is where we securitize the transactional risk of
electronic commerce.**
There is one potential fly in this ointment, and I do not intend to dwell
on it, but I cannot get this far and not mention the threat to strong
security apparati of having them undermined by key escrow. Corporate
policies and laws alike have always been defined in a territorial way that
relies on clearly identifiable borders, physical locations where the
policy or the law come to an end. But in the electronic world borders are
meaningless. In some sense, sovereignty, based as it was on the idea of a
border, is less meaningful now than for some centuries. In its place is a
different kind of sovereignty, because the only borders in an electronic
world are cryptographic ones. As such, the debate over who may or may not
have a key known only to themselves is a proxy discussion for who may or
may not have sovereignty within a cryptographically defined space.
There are hard questions yet to answer. Compromised keys are revoked
effective not to the moment of suspicion of compromise but rather
retroactively to the last known time when the key was safe. In the case of
escrow, should not a key's owner retroactively revoke it to the moment of
its seizure from escrow should the owner later discover that it has been
so seized? Or if a revoked key is only revoked by the action of the
certifying authority signing a revocation notice in a special key, can
that revocation-signing key itself ever be revoked? If it could, would
that not invalidate (reverse) any revocations signed in it and what does
that mean? I only offer these so that you do not equate my argument about
the near-inevitability of investment in public key technology and
digital-signature-dependent activities with some presumed infallibility of
the technology or our understanding of it. These questions will be
settled one way or another, but they remain open as we speak here today,
and there is money to be made.
I have tried to lay out my estimation on which way the tide is running and
which moon's gravity matters. I could be completely wrong, or merely
overstating what my biases bring me, but I think not. I think that just as
the best estimate of tomorrow's weather is today's, the best estimate of
how the Internet and the financial behemoths will interact is for the
Internet to be driven, as a side effect, by the cost-reduction and
profit-incented strategies of those financial behemoths. They already
transcend national boundaries and their investment decisions do run the
world. Were this to get enough investment, it might make security a
solved problem at least as I define "solved" to mean "consistent with risk
management in the insurance style." Since that would collapse the market
for novel security add-ons, I strongly suggest that as you prepare your
business plans you figure out how to be, as Tom Lehrer would say, a doctor
specializing in diseases of the rich.
This is a very exciting time and it is a privilege to be a part of it.
When we are all relics in rocking chairs, we will still know that we were
present at the creation. I know that I will count myself particularly
lucky, including for your close attention these past few minutes.
Thank you for the honor of speaking with you.
END
-o-
Subscribe: mail majordomo@repsec.com with "subscribe isn".
Today's ISN Sponsor: Repent Security Incorporated [www.repsec.com]
Received on Thu Nov 12 09:22:09 1998